The issue of ballistic missile defense, more recently known as National Missile Defense (NMD) or Star Wars, is in the news now more than it has been in years, thanks to President Bush's current effort to abrogate the Anti-Ballistic Missile Defense treaty and push ahead with developing and deploying a system.

In the last few years, the United States has increased annual spending on ballistic missile defense to 1.9 billion dollars (as per FY01 Budget Request [ http://www.cdi.org/issues/nukef&f/nukecost.html ]); we have now spent a total of some $60 billion developing a system. At the same time, in this post-cold-war era, the official justification has been scaled far back; we no longer expect it to protect us against an attack by a powerful adversary like the Soviet Union, but only against an obviously-irrational attack by a "rogue state" like North Korea. Is this money well spent? What are the odds we'll end up with a system that really protects us? The vast majority of press coverage I've seen focuses on questions like the difficulty of hitting an attacking missile with whatever technology we plan to hit it with, and what that technology should be--lasers, missiles of our own, whatever. But in my view and that of many other computer experts, the important question is software.

I have a Ph.D. in computer science, but I've spent most of my 30-year career as a programmer, developing complex software, not as a researcher or professor. I base my opinions on both theory and experience. Even based on the Pentagon's own estimates, the NMD software system would be the most complex piece of software ever developed. Furthermore, it would be software of an unusually difficult type. It would have to operate in "real time"--that is, to complete its calculations in a period of time determined by outside forces. And it would operate in a very hostile environment: in addition to the problems all software faces, it would have human adversaries trying to defeat it (perhaps by repeatedly changing the design of their missiles, perhaps by spies feeding them copies of the program code).

It has been said that, ordinarily, software engineers are programming Murphy's computer (as in Murphy's Law: anything that can go wrong, will). In this case, they would be programming something even worse, the devil's computer! And how could this software be debugged? Simulation is routinely mentioned as the main way, and indeed simulation would undoubtedly find many problems. Well, simulation is great for finding the problems you suspect as well as for testing the solutions you think you have. But it's not much good for finding problems that never occurred to you, and in a system of the overwhelming complexity of NMD, there would surely be many of those.

But wait. Computers these days are absolutely dazzling. If we can build machines that would have been considered supercomputers only a few years ago and that sell for $3,000, can ballistic-missile defense software really be beyond us? Yes, it can. Year after year, there's been incredible progress in computer hardware, but not much in software, and almost none in "reliable" software. We now have on our desktop computers of blinding speed that crash as often as the dinosaurs of 20 years ago.

And even if we could be confident the NMD system would work flawlessly, it's very doubtful that it would accomplish anything. Why should those North Koreans, deranged though they might be, send expensive missiles to destroy our cities when they could smuggle their nukes into harbors on ships? Or send them over in cheap pilotless planes? With respect to the latter, in 1998, a group of meteorologists led by Tad McGreer showed that GPS-equipped robot planes built by amateurs for about $10,000 each can cross the Atlantic Ocean. Such planes--low-tech cruise missiles powered by readily-available model aircraft motors--could fly below our radar defenses and deliver payloads like a small atomic bomb or a quantity of homemade nerve gas.

Finally, even a primitive system could act in an offensive way. In recent years, Russian strike capabilities have declined considerably, and they continue to decline. At some point, a baby Star Wars could be good enough to knock down most or all of a weak Russian retaliation to a US first strike, thus making it very destabilizing. This was a big problem with the original Star Wars. Of course, the Chinese and Korean ICBM programs, if they improve at any reasonable rate, will rapidly make NMD useless anyway. A cynic might argue that Baby Star Wars is really just a way to continue giving as much money as possible to military contractors, and to avoid confronting our bad habit of handling international relations problems via force.

The proposed system would be a colossal waste of money that we can ill afford, even in these days of prosperity. This is especially true because it distracts us from the need to improve international relations, an incomparably more effective way to protect our country. Many years ago, Jerome Wiesner, then president of MIT, argued that the threat of nuclear war is one of a class of "no technical solution" problems, and that attempts to avoid it via technology would in fact make the problem worse. Ballistic missile defense is as a good a demonstration of his point as could be imagined.

The Star Wars version of the Maginot Line is no better an idea than the World War II version.

What's inside...